For cancer treatments, clinical trials have a 5% likelihood of success in moving from preclinical trial to FDA approval. This success rate has remained largely unchanged since the early 2000s. Owkin and the team at Sorbonne University and Assistance Publique - Hôpitaux de Paris (AP-HP) are looking to change that by learning more about cancer patients, through artificial intelligence.
Félix Balazard, Director of Optimized Development, Owkin, says:
This collaboration was born out of an alignment of vision: Owkin, Sorbonne University and AP-HP see great potential for patients through the application of AI.
The teams will focus initially on pancreatic cancer, a cancer that is frequently diagnosed at an advanced stage, thereby reducing the chances of recovery, since symptoms typically only appear after it has spread to other organs.
Magali Svrcek, Project Leader and Professor of Pathological Anatomy and Cytology at Sorbonne University, says:
Despite significant investments in research and development, all recent trials have proven disappointing, leaving patients without effective treatment.
A new approach
Owkin will use AI to uncover new insight from multimodal patient data, helping scientists and physicians learn more about the biology of cancer patients and better design clinical trials.
AP-HP are no strangers to harnessing technology in the pursuit of a deeper understanding of disease - they will soon be deploying digital pathology across the hospital, allowing them to extract previously hidden detail on tumors, details that can measure if tumors are responding to treatment, or reveal what molecular abnormalities they contain.
Along with the hilogy slides generated through digital pathology and electronic medical records, data extracted from liquid biopsies will be used. This emerging non-invasive technique allows scientists to trace cancer progression through blood sample analysis.
Data from liquid biopsies (circulating tumor cells and DNA) seem to be extremely relevant in pancreatic cancer to help refine prognostic algorithms and therefore better select patients for a given treatment.
Optimizing clinical development through AI
The probability of clinical trial success is heavily tied to the distinct biology of the patients recruited. Félix explains:
Pancreatic cancer, like all cancers, is very complex and presents itself differently in each patient.
AI models are very good at uncovering the prognostic similarities within a group of patients. These insights can fuel methods like covariate adjustment - a statistical approach in the analysis of randomised controlled trials - to improve clinical trials. This is particularly impactful for aggressive cancers such as pancreatic cancer as we have shown in our recent publication. We have engaged the EMA on the use of deep learning histological covariates and they have issued a Letter of Support on the topic.
We hope the results from AI-TRIOMPH will bolster that support, further demonstrating the effectiveness of these AI techniques to regulators.
Following pancreatic cancer, the collaboration will expand to focus on two additional cancers, esogastric cancer and refractory thyroid cancers.
The project is supported by a RHU grant from ANR and is aligned with the FRANCE 2030 strategy on artificial intelligence. “The idea, ultimately, is to build a hub of digital medical data of different types around cancer (“Cancer Intelligence Hub”) which would be usable by the national and international scientific community.”